Hidden poison exons in the human brain reveal a new route into neurological disease

Deep inside the human genome, a hidden layer of genetic regulation appears to be far more important for the brain than scientists have appreciated. A new study published in Genome Biology by a team at Massachusetts General Hospital and Harvard Medical School maps, on a genome-wide scale, the locations and behavior of so-called poison exons — short stretches of DNA that, when spliced into a messenger RNA molecule, sabotage the very gene that produced it. The research, led by Paolo Pigini and Huilin Xu as co-first authors, with Dadi Gao and Elisabetta Morini as corresponding authors, identifies nearly ten thousand of these self-destruct switches in the human genome and shows that hundreds of them are uniquely active in the brain, where they may hold the keys to a wide range of neurological disorders.

Poison exons are a striking example of evolutionary ingenuity. They are highly conserved exon sequences whose inclusion in a mature transcript introduces a premature termination codon, an early stop signal that the cell recognizes as a mistake. Rather than translating the flawed message into a truncated and potentially harmful protein, the cell destroys the transcript through a quality-control pathway known as nonsense-mediated decay, or NMD. In effect, a poison exon acts as a molecular volume knob: when it is spliced in, the gene is silenced; when it is spliced out, the gene’s protein is produced. Because this mechanism allows cells to fine-tune protein levels rapidly and reversibly, without changing the underlying DNA, poison exons have long been suspected to play a major role in post-transcriptional regulation. What has been missing is a systematic, transcriptome-wide inventory of where they are and what they do.

The new study addresses that gap with unusual breadth. The researchers performed a comparative analysis of 954 eukaryotic transcriptomes spanning a wide range of species, asking how frequently each organism produces RNA isoforms targeted by nonsense-mediated decay. The answer was unambiguous: Homo sapiens stands out as the species with the highest enrichment of NMD-targeted isoforms, suggesting that this form of regulation became especially elaborate in the human lineage. Building on that observation, the team systematically predicted and annotated splicing events capable of generating premature termination codons across the human genome, cataloging 9,814 poison exons in total — a dramatic expansion over the fragmentary annotations that existed before.

With the catalog in hand, the researchers turned to two of the most valuable resources in human genomics: GTEx, which profiles gene expression across dozens of tissues from hundreds of donors, and BrainSpan, which charts transcriptomic changes across human brain development from prenatal stages into adulthood. By mining RNA sequencing data from these repositories, they mapped how poison exon inclusion varies across the body and across the lifespan. The results were striking. The team identified 83 poison exons found uniquely in the human brain, 973 that show brain-specific differential splicing compared with other tissues, and 1,284 whose splicing changes significantly during brain development. In other words, a substantial fraction of the genome’s poison exon repertoire is deployed in neural tissue, and its deployment shifts as the brain matures — a pattern consistent with a regulatory system tailored to the demanding and precisely timed business of building and maintaining neurons.

What kinds of genes carry these brain-relevant poison exons? The functional and pathway enrichment analyses performed by the team point toward processes at the heart of neural biology, including synaptic function and neuronal development, and toward disease categories dominated by neurological conditions. That connection moved from correlation to mechanism when the researchers cross-referenced their poison exon map with ClinVar, a public archive of clinically observed genetic variants, and with SpliceAI, a deep-learning tool that predicts how DNA variants alter splicing patterns. The screen revealed 1,625 pathogenic variants associated with neurological disorders that are predicted to disrupt the splicing of 743 poison exons. Each such variant, in principle, could tilt a cellular volume knob in the wrong direction — forcing a poison exon into transcripts that should exclude it, or removing one from transcripts that need it, thereby raising or lowering the level of a critical neural protein.

Crucially, the team did not stop at prediction. They functionally validated the impact of a subset of these variants using CRISPR prime editing in human cells, a precision genome-engineering technique that can introduce exact genetic changes without making double-strand breaks in DNA. The experiments confirmed that the predicted splicing defects were real: engineered variants changed poison exon inclusion in the predicted direction, providing direct experimental support for the computational pipeline and for the broader idea that poison exon mis-splicing is a genuine disease mechanism rather than a bioinformatic curiosity.

The most detailed case study centered on a variant with a known clinical consequence. The intronic NDUFAF6 c.420+784C>T variant is associated with Leigh syndrome, a devastating mitochondrial disorder that typically strikes infants and progressively damages the brain and other energy-hungry organs. NDUFAF6 encodes an assembly factor required for building mitochondrial complex I, the molecular machine that generates much of the cell’s energy supply. Using induced pluripotent stem cell-derived neurons — i3N neurons, a rapid differentiation system that produces neuronal-like cells from patient or engineered stem cells — the researchers showed that the variant causes increased inclusion of a poison exon in the NDUFAF6 transcript. The consequence was a significant reduction in NDUFAF6 protein levels, and downstream of that, impaired mitochondrial membrane potential compared with control neurons. The chain of causation was thus traced end to end: a single DNA letter change, deep inside an intron, rewires splicing to insert a premature stop codon, the cell destroys the transcript, protein levels fall, and mitochondria malfunction — the cellular signature of Leigh syndrome.

The implications of this work extend well beyond a single gene or disease. Because poison exons act as tunable regulators of protein abundance, they represent an unusually attractive class of therapeutic targets. Antisense oligonucleotides — short synthetic RNA molecules that can be designed to block or promote specific splicing events — are already approved for other splicing-based disorders, and the study’s authors note that their own patent portfolio includes work on RNA splicing modulation. A poison exon that is aberrantly included in a patient’s neurons could, in principle, be skipped by a designed oligonucleotide, restoring normal protein production without ever touching the genome. Conversely, prime editing or base editing strategies could correct the underlying splicing variants at their source, an approach the team has explored for familial dysautonomia, another splicing-driven neurological disease. The new catalog of 9,814 poison exons, together with the list of 743 whose splicing is predicted to be disrupted by pathogenic variants, effectively hands the therapeutic community a prioritized target list.

The study also reframes how scientists should think about the enormous fraction of disease-causing mutations that lie outside protein-coding regions. Variants in introns and other non-coding DNA have often been difficult to interpret, but the NDUFAF6 case demonstrates that a deep intronic change can be every bit as pathogenic as a mutation in a coding sequence — provided one knows to look for its effect on splicing. Tools like SpliceAI, applied systematically as they were here to more than 236,000 ClinVar variants, can surface such effects at scale, and the authors compared SpliceAI’s performance against another splicing predictor, Pangolin, to strengthen their calls. As diagnostic sequencing expands, incorporating poison exon-aware splicing prediction into variant interpretation could resolve cases that currently go undiagnosed, particularly in neurodevelopmental and neurodegenerative conditions where the genetic basis remains elusive for many patients.

There are caveats and open questions, as with any ambitious mapping effort. The study’s tissue and developmental analyses rest on bulk RNA sequencing data, which averages signals across the many cell types of the brain, and the functional validations, while rigorous, cover a subset of the predicted events. The authors acknowledge that the field’s understanding of poison exon regulation — which splicing factors control which switches, and in response to what signals — is still in its infancy. Yet the scale of the new resource, the evolutionary signal that humans are the most NMD-enriched species examined, and the demonstration that a single poison exon mis-splicing event can recapitulate the mitochondrial pathology of Leigh syndrome together make a compelling case. Poison exons, once a footnote of RNA biology, now look like a central chapter in the story of how the human brain regulates its genes — and a promising frontier for treating the diseases that arise when that regulation fails.

Subject of Research: Poison exon splicing and its role in gene regulation and neurological disorders in the human brain

Article Title: Poison exon splicing in the human brain provides a new paradigm for understanding and targeting neurological disorders

Article References: Poison exon splicing in the human brain provides a new paradigm for understanding and targeting neurological disorders. (n.d.). https://doi.org/10.1186/s13059-026-04292-8

Image Credits: AI Generated

DOI: 10.1186/s13059-026-04292-8

Keywords: poison exons, alternative splicing, nonsense-mediated decay, RNA splicing, human brain, neurological disorders, Leigh syndrome, NDUFAF6, SpliceAI, CRISPR prime editing, GTEx, BrainSpan